Back to skill

Security audit

arena

Security checks for vulnerabilities and agentic risk

Overview

Arena’s agent workflow is well scoped, but ordinary use can automatically install and execute unreviewed updates.

Before use, disable automatic updates or use a Git installation pinned to a reviewed commit. Otherwise, invoking Arena can replace its instructions and run new upstream shell code with the agent’s existing permissions. ClawHub installations also lose local edits to the three updated files.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description promises update checks, but the instructions require installing changes from a mutable public main branch and immediately adopting the downloaded instructions for the current task, without specifying integrity verification or a separate approval gate. An upstream compromise could therefore replace trusted workflow instructions and influence subsequent tool use within the session's available permissions. The updater source was not provided, so the additional claim that it executes a newly downloaded updater script cannot be verified from this file.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The declared update check automatically installs upstream changes and executes the newly downloaded updater.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The updater replaces local skill files from the mutable upstream main branch and immediately executes the downloaded update.sh without approval or independent integrity verification. An attacker who compromises the upstream repository or its publishing credentials could therefore execute arbitrary commands with the invoking agent's permissions; checking the repository URL and rejecting source symlinks do not authenticate the downloaded code. The skill's advertised update check makes this more dangerous because normal invocation can become installation and execution of previously unreviewed code.

Content

No source excerpt is available for this finding.